Episodic memory governs choices: An RNN-based reinforcement learning model for decision-making task.

Area of Science:

• Computational Neuroscience
• Cognitive Science
• Artificial Intelligence

Background:

• Studying cognitive function typically involves recording neural activity during behavioral tasks, but often fails to capture all relevant neurons.
• A significant challenge in neuroscience is understanding how the brain selects episodic memories to guide future decisions.

Purpose of the Study:

• To develop a computational framework for studying cognitive functions, overcoming limitations of current neural recording methods.
• To investigate the role of episodic memory retrieval in decision-making and learning speed.

Main Methods:

• Developed a Recurrent Neural Network (RNN)-based Actor-Critic framework trained using reinforcement learning (RL).
• The framework was trained on tasks analogous to primate decision-making tasks.
• Behavioral experiments were conducted on the computational framework to explore memory selection strategies.

Main Results:

• The trained model reproduced key features of neural activity and behavioral properties observed in animal experiments.
• The framework demonstrated that retrieving salient episodic memories, rather than common ones, reduced deliberation time.
• Prioritizing salient events facilitated faster learning of decision-making strategies.

Conclusions:

• The RNN-based Actor-Critic framework serves as a valuable computational platform for exploring cognitive functions.
• Salient events stored in the hippocampus appear to be prioritized for propagating reward information, enhancing learning efficiency.
• This research offers insights into the neural mechanisms underlying memory-guided decision-making.